Professional Documents
Culture Documents
0241 to TLP
Franz Berto
University of St Andrews and University of Amsterdam
Abstract
Tractatus 4.024 inspired the dominant semantics of our time: truth-
conditional semantics. This has focused on possible worlds: the con-
tent of p is the set of worlds where p is true. It has become increas-
ingly clear that such an account is, at best, defective: we need an
‘independent factor in meaning, constrained but not determined by
truth-conditions’ (Yablo 2014, p. 2), because sentences can be differ-
ently true at the same possible worlds. I suggest a missing comment
which, had it been included in the Tractatus, would have helped se-
mantics get this right from the start. This is my own 4.0241: ‘Knowing
what is the case if a sentence is true is knowing its ways of being true’:
knowing a sentence’s truth possibilities and what we now call its topic,
or subject matter. I show that the famous ‘fundamental thought’ that
‘the “logical constants” do not represent’ (4.0312) can be understood
in terms of ways-based views of meaning. Such views also help with
puzzling claims like 5.122: ‘If p follows from q, the sense of “p” is
contained in the that of “q”’, which are compatible with a conception
of entailment combining truth-preservation with the preservation of
topicality, or of ways of being true.
1
Tractatus logico-philosophicus (TLP)’s section1 4.024 has been at the core of
the dominant semantics of 20th Century: truth-conditional semantics. This
has focused on truth-at-possible-worlds conditions. The meaning or content
of a sentence is given by its intension: a function from possible worlds to
truth values or, equivalently, the set of possible worlds where the sentence is
true. Call this Standard Possible Worlds Semantics (SPWS).
Possible worlds don’t show up explicitly in the TLP. Let’s thus start by
laying on the table some textbook Tractarian semantics – by which I mean: a
minimal account, which doesn’t get into big interpretative issues.2 Say that
an atomic sentence is one that doesn’t have further sentences as its syntactic
constituents. Such a sentence is meaningful by being a picture of a state of
affairs (4.032): a possible configuration or combination of objects, perhaps
of objects and properties/relations (2.01, 2.031-2.032, 2.202). The state
of affairs is the sense of the sentence qua picture (2.221, 3.13), what we
know by understanding it (4.021). States of affairs can obtain, or fail to
obtain. Obtaining states of affairs are facts (The TLP’s terminology is not
very uniform here, but let’s not quibble over this either). By obtaining, a
state of affairs makes true the sentence it is the sense of (4.25). The world
is the totality of facts, that is, of obtaining states of affairs (1-1.2, 2.04-
2.06, 2.063). Thus, all the true atomic sentences taken together make for
a complete description of the world (4.26). Complex sentences are truth-
functional compounds of atomic sentences (5): their senses consist of the
combinations of the obtaining and non-obtaining of states of affairs pictured
by their atomic constituents, which make them true (4.2, 4.4-4.41, 4.431).
How did we get from (something like) this to SPWS? Carnap is usually
taken as the main responsible. When introducing his method of intension and
extension in Meaning and Necessity, he claims that ‘some ideas of Wittgen-
stein were the starting-point for the development of this method’ (Carnap,
1947, 9). Notoriously, he talked of state descriptions, not possible worlds,
but a state description is ‘a class of sentences [...] which contains for ev-
1
I will call ‘sections’ the numbered sentences, or groups thereof, composing the TLP
(Wittgenstein, 1921/22); these are often called ‘propositions’, but I find the terminology
a bit confusing. I will use ‘sentence’ as short for ‘declarative sentence’: a linguistic config-
uration that can be true or false. The TLP uses ‘Satz’ for this, and the Ogden translation
has ‘proposition’; I will only leave the word with this meaning when directly quoting the
TLP. I myself will use ‘proposition’ or ‘(propositional) content’ for the meaning or content
of a declarative sentence.
2
Like, What exactly is an elementary or atomic sentence? Does it include predicative
terms? What’s an atomic fact? How do the objects that constitute an atomic fact hang
together? Are properties and relations objects, too? What are Tractarian objects like, by
the way? – and so on. There are many excellent guides to the TLP, giving overviews of
the debates on these issues. I recommend especially Frascolla (2007) and White (2006).
2
ery atomic sentence either this sentence or its negation, but not both, and
no other sentences’ (Ibid.). Ignoring various complications, one may take
them as close enough to complete descriptions of possible worlds – or, as
the worlds themselves, in a linguistic ersatz (Lewis, 1986, Ch. 3) account
of them (Carnap claims, a bit misleadingly, that ‘the state-descriptions rep-
resent Leibniz’ possible worlds or Wittgenstein’s possible states of affairs’,
Ibid). Equivalently, one can take the totality of possible states of affairs,
and form maximally consistent combinations of them. These will be, in a
combinatorial setting, the possible worlds. Let their collection be W . The
actual world is but the totality of states of affairs that obtain. And then we
have it: the intension of a sentence p, giving its propositional content (or
so-called UCLA proposition), is a function Ip : W → {T, F } mapping each
w ∈ W to its extension: truth, if p is true at w; falsity otherwise.
2 + 2 = 4.
These are true at the same worlds (all of them, if mathematical necessity
is unrestricted) but they don’t seem to mean or say the same thing: only one
is about triangles, and made true by how they are. One may quibble over
the use of necessary truths, but take an example made famous by Hempel:
These are (classically) logically equivalent, thus true at the same worlds,
and arguably contingent, but they cannot quite mean the same, for different
pieces of evidence confirm them. Only one is about ravens, and made true
by how ravens happen to be.
3
According to Yablo, aboutness is the missing ingredient in SPWS. This is
‘the relation that meaningful items bear to whatever it is that they are on or
of or that they address or concern’ (Yablo, 2014, 1): their subject matter or,
as I shall also call it, their topic. I’ll say something on what topics or subject
matters are, or could be, in a minute. Here’s what they are not – or, cannot
easily reduce to: the things referred to in a sentence. Topicality also has to
do with what is said about those things. ‘Dog bites man’ and ‘Man bites
dog’ (Yablo, 2014, 34) involve the same things: dog, man, perhaps biting –
but they don’t say the same.
The main alternative to SPWS for an account of propositional content
comes from structured propositions. Can’t they get it right? In the so-
called Russellian propositions view, the content of ‘John kisses Mary’ is a
structured object involving John, Mary, and the relation of kissing, in this
order. Ordering can tell it apart from the content of ‘Mary kisses John’.
There are several problems with structured propositions as an account of
content and same-saying at the right level of semantic fine-grainedness. I
cannot get into them here but, for an insightful discussion, one can read
(Ripley, 2012). I’ll just mention one issue, raised in (Yablo, 2014, 1), for it
will be important in the following: in the Russellian account, ‘Mary does not
kiss John’ differs in content from ‘Mary kisses John’, in that the content of
the former includes a component, not, which the content of the latter lacks.
Also, the content of ‘Mary kisses John and Paul is jealous’ includes and as a
component. This seems wrong for subject matters: there’s no sense in which
‘Mary does not kiss John’ talks about negation, or ‘Mary kisses John and
Paul is jealous’ talks about conjunction.
Work on subject matter and ways of being true has been flourishing
lately. This has revitalized a semantic tradition to some extent alternative
to SPWS, and focused on a mereological view of meaning, whereby contents
can be taken as having parts, as including other contents, as capable of being
fused into wholes which inherit the proper features from the parts. Ideas in
this ballpark can be found already in Parry (1933), Ryle (1933), Goodman
(1961), Perry (1989). Ways of being true have more recently been understood
in terms of partial content (Humberstone, 2000), world-partitions (Lewis,
1988), world-divisions (Yablo, 2014), truthmakers (Fine, 2014, 2015), and
more (Hawke, 2017). The overall aim is always the same: giving a better-
than-SPWS account of when two sentences say the same thing, and when
they say different things.
I don’t think that the semantic ideas included in the TLP lead inevitably
to SPWS. I’m not a specialist of Wittgenstein – only an admirer of the TLP
since I was an undergrad. Still, it seems to me that there are, in it, traces
4
of an alternative, mereological view of content. Before proposing, in two
Sections, a ‘missing gloss’ to TLP 4.024 which, had it been there, would
have helped 20th Century semantics get things right from the start, in the
next Section I will dig into a place in the TLP, where such traces seem to
show up: the Tractarian account of logical consequence.
5
Ramsey’s way of making sense of sense containment has it that, when by
asserting p we happen to assert q, the sense of q has that of p as a part. As
an intuitive (‘conforming to ordinary usage’) example, he points at the sense
of ‘p ∧ q”s including that of p. Because p ∧ q entails q according to SPWS,
this does not tell sense containment apart from SPWS logical consequence.
But there’s another intuitive example in the vicinity, which does: p entails
p ∨ q according to SPWS, but surely ‘p ∨ q’ can be partly about something
p is not about, namely whatever q is about. Flag both of these entailments,
for they will be important in the following.
Max Black is more sanguine than Ramsey: he calls the idea of sense
containment ‘an impediment to clarity’, and glosses on 5.122 by speaking
of ‘a peculiar (and possibly unfortunate) use of this word [scil. ‘contained’]’
(Black, 1964, 251-2). Fogelin (1976) says that sense containment is no more
than a metaphor.
Can we make sense of sense-containment, asks Negro? Negro himself, and
Frascolla (2007), have interesting proposals (differing from each other in ways
I won’t get into here), that pivot on looking, rather than at truth-grounds,
at their complements, namely, falsity-grounds, in a negative or exclusionary
view of content (Rumfitt, 2008). We can then explain the reversed inclusion
direction in sense containment as some kind of inclusion of falsity-grounds.
Now I want to leave Wittgensteinian exegesis behind, though, and broaden
the view a bit. There is a less-then-metaphorical tradition in logic and se-
mantics – variously stretching back at least to Kantian ideas on analyticity
– of looking for content-preserving entailment relations: relations that hold
between meaningful q and p only when p introduces no content alien to that
of q. Among such relations are those labeled as ‘tautological entailment’
(Van Fraassen, 1969), ‘analytic containment’ (Angell, 1977), ‘analytic impli-
cation’ (Parry, 1933; Fine, 1986; Ferguson, 2014). I will sketch two ways
to approach content-containment and ways-based semantics in the coming
Section, after I’ve added my proposed gloss to 4.024.
4.0241* Knowing what is the case if a sentence is true is knowing its ways
of being true.
6
How do we understand such ways of being true? One way to have a ways-
based account of propositional content is found in Kit Fine’s works, e.g.,
(Fine, 2014, 2015). Fine’s core idea is very Tractarian in spirit: sentences
are made true by states of affairs or situations, rather than by whole possible
worlds. His semantics has a set with a partial ordering on it, hS, ≤i. Each
s ∈ S is a situation, or state of affairs, or a configuration of objects, or of
objects and properties. States of affairs are things that can obtain or fail,
and are chunks of reality that ground the truth and falsity of sentences, so
we’re not tremendously far from TLP’s own states of affairs.
We are far from possible worlds, however – though one can recover a
certain account of worlds by taking them as limit constructions out of states.
The ordering s1 ≤ s2 may be read in a metaphysically loaded way: ‘s1 is a real
part of s2 ’, thus, states of affairs can be literally included in larger ones (this
works well mainly with states of affairs involving concrete entities: the state
consisting of St Andrews’ being in Scotland is included in the state consisting
of St Andrews’ being in the UK). I favor a less metaphysically loaded reading
of ‘≤’ in terms of information: ‘s2 preserves all the information in s1 ’, or ‘s2
supports the truth of anything whose truth is supported by s1 ’, or so. Then
what matters about states of affairs, for semantics, is the information they
encode or support.
Either way, worlds would be things which are maximal with respect to
≤: when w1 and w2 are worlds, the only way for ‘w1 ≤ w2 ’ to hold is for
w2 to be w1 . Finean states are not like that: like the situations of Barwise
and Perry (1983)’s situation semantics, they are partial and relevant for the
things whose truth they support. And it’s states, not worlds, which are at the
core of Kit Fine’s semantics – just as they are at the core of the Tractarian
theory of representation: it’s of facts, i.e., (obtaining) states of affairs, that
we make pictures of (2.1). Pictures are facts (2.141): the facts consisting
in their elements’ being in such-and-such relations with one another (2.14).
The elements of the picture stand for (vertreten) the objects which are the
constituents of the pictured state of affairs (2.131) – and so on, as per the
well-known (though variously interpreted) pictorial theory of representation.
Thoughts are logical pictures, 3, and sentences are the expression of thoughts,
3.1. The sense of a sentence – its content – is the state of affairs pictured
by the sentence. A sentence is about the state of affairs that makes for its
sense (this is what the sentence shows, according to 4.022), and says that
things are such-and-so with respect to it: ‘aRb’ is about the state of affairs
consisting in the thing named by ‘a’ ’s being R-related to the thing named
by ‘b’, and that ‘a’ is in a certain relation with ‘b’ says that aRb (3.1432).
What I understand when I grasp the content of p, is its the state of affairs p
represents (4.021).
7
I will not get into the details of Fine’s state-based semantics, however,
for I favor a different way of having a way-based semantics. This comes from
two-component (2C) accounts of content, which are more friendly to worlds.
One account of this kind has been developed in great detail in (Yablo, 2014),
but I’ll stick to a more abstract 2C setting I’m working on, together with
some friends (Berto, 2018a,b; Hawke, 2017). The content of a sentence p is
the thick proposition it expresses, [[p]]. This has two components: (1) p’s
intension, |p| ⊆ W , its truth set as per SPWS (the thin UCLA proposition),
and (2) p’s topic or subject matter, t(p). Overall, [[p]] = h|p|, t(p)i. p and q
say the same, that is, express the same thick proposition, when (1) they are
true at the same worlds, and (2) they have the same topic. Given that topics
can have parts, we also get a natural view of content inclusion: the content
of p includes that of q when (i) any p-world is a q-world and (ii) the topic of
q is part of that of p. When both (i) and (ii) obtain, we claim that p ‘says
more than (or, at least as much as)’ q.
Two natural questions now are, Why two components? What are these
topics or subject matters? As for the first one, our understanding of propo-
sitional contents can be naturally seen as involving two elements. Now I
attempt a further envisaged addition to the TLP – a subsection glossing on
my proposed 4.0241*, and pushing more in the 2C direction:
4.02411* One knows how a sentence is true by knowing its truth possibilities
and what it is about.
One understands what p means or says, when one understands (1) what
the sentence speaks about, its topic or subject matter, and (2) what it says
about that topic or subject matter. One understands ‘La neige est blanche’
as soon as one knows that (1) it speaks about the color of snow, how snow is
like, snow’s whiteness, etc. – and, one knows that (2) it says that things are
such-and-so with respect to that subject matter, that is, that snow is white.
In a plain state-based account à la Fine, the two components are taken
care of by one kind of things: the states. A state of affairs verifying p is both
what p is about, and what truthmakes it. But we may have tentative reasons
for splitting the two components. It seems that people can sometimes have a
partial understanding of the meaning of a sentence by grasping only its truth
conditions, or only its topic. William III could have understood the truth
conditions of ‘Either France will get into a nuclear war with England, or not’:
he’d know, by only looking at its logical form, that the sentence would be
true no matter what. He couldn’t have grasped what the sentence is about,
for he had no idea of what a nuclear war is. William III could have grasped
the topic of ‘Louis XIV is bald’ (say, Louis’ baldness) without knowing under
8
which conditions the sentence would be true (just in case Louis had n hairs
– for what n? A range between some n and some m? Which n and m then?
A fuzzy range? etc.).
That we split components doesn’t mean that we must treat them as
conceptually or metaphysically irreducible to each other. Here our second
question comes in: What are topics, or subject matters? One may take
them, again, as truthmakers or states of affairs, and reduce worlds to them by
taking worlds as maximally informative states, or constructions out of states.
But one may also go the other way around. This was the way pursued in
Lewis (1988)’s seminal work on subject matters. Understand subject matters
starting from questions: the subject matter of p (in context c), is given via the
question or questions p can be taken as answering to (in c). This determines
a partition of the set of worlds: w1 and w2 end up in the same cell when
they agree on the answer. ‘The number of stars is eight’ has as its topic the
number of stars. It can be taken as an answer to the question, How many
stars are there? This splits and groups the worlds depending on how they
answer: all the zero-star worlds end up in one cell; all the one-star worlds
end up in another; etc.
Yablo (2014) proposes to generalize Lewisian partitions, thus equivalence
relations, to divisions: sets of worlds that collectively cover the modal space,
but which allow overlap, determined by similarity relations which are reflexive
and symmetric, but not transitive. Some questions have more than one
correct answer (‘Where can I find a B&B in Kirchberg?’), so a world can be
in more than one cell with respect to them. In any case, subject matters
reduce to ways of splitting and grouping worlds into sets.
Before I give you more 2C details, I need to talk of Wittgenstein’s fun-
damental thought.
9
We saw that there is something wrong in the claim that ‘Snow is not white’ is
about negation. ‘Snow is not white’ must be about whatever ‘Snow is white’
is about: that may be snow’s color, the whiteness of snow, etc. Of course, the
two sentences say opposite things about that topic, but the topic they address
is the same. In general, ‘¬p’ must be exactly about what p is about. We can
phrase this first guiding principle for a propositional recursion on topics in
2C-terms: t(¬p) = t(p). Thus, also, t(¬¬p) = t(p): remember Frege on the
Sinn-preservation of double negation, a view endorsed by Wittgenstein. (Of
course, [[p]] 6= [[¬p]] because |p| is the complement of |¬p|.) In Tractarian
terms:
4.0621 That, however, the signs “p”and “q” can say the same thing is im-
portant, for it shows that the sign “∼” corresponds to nothing in reality.
That negation occurs in a proposition, is no characteristic of its sense
(∼∼ p = p). [...]
5.44 [...] And if there was an object called “∼”, then “∼∼ p” would have
to say something other than “p”. For the one proposition would then
treat of ∼, the other would not.
10
• t : LAT → T assigns a topic t(p) to each atom p.
Out of fusion we can define what it means that topic x is part of topic y,
x ≤ y =df x ⊕ y = y – making of parthood a partial ordering (with the strict
ordering, <, defined from the nonstrict ≤, the usual way). v is extended to
the whole L via the usual recursive clauses assigning a truth set |A| to each
formula A of L:
• |¬A| = W − |A|
• |A ∧ B| = |A| ∩ |B|
• |A ∨ B| = |A| ∪ |B|
• t(¬A) = t(A)
11
(say, A B: ‘B follows from A’, or ‘A entails B’), that embeds topic-
containment. We now know what it means, in a 2C-setting, that one thick
proposition includes another: [[A]] includes [[B]] (let’s write ‘[[A]] [[B]]’,
that is, h|A|, t(A)i h|B|, t(B)i) when (1) |A| ⊆ |B|, all the A-worlds are
B-worlds, and (2) t(B) ≤ t(A), that is, the topic or subject matter of B
is included in that of A. We can interpret in 2C terms ‘A proposition af-
firms every proposition that follows from it’ (5.124), and ‘If one proposition
follows from another, then the latter says more than the former, and the
former says less than the latter’ (5.14): we have precise characterizations
of same-saying, saying strictly more, saying strictly less, saying at least as
much. In particular, we claim that A M B (A entails B in model M) just
in case [[A]] [[B]] in M, that is, A says at least as much as B there: while
A’s truth set involves no more worlds, its subject matter is at least as big as
that of B. And we say that A B when A M B for all M.
12
intuitively grasped the general case as correct, [this] was thus left completely
unexplained in the Tractatus’ (Negro, 2017, 6).
In the TLP, A entails A ∨ B in the 5.12 sense of entailment: all the
truth-grounds of the former are truth-grounds of the latter or, as we would
say today, any possible world in any interpretation or model making A true
will also make A ∨ B true. But, unlike A ∧ B A, this entailment is not
topic-preserving, for B may break the boundaries of A’s subject matter: it
is not the case that, in general, whatever ‘A ∨ B’ says was already said by
A. These have been taken as paradigmatic cases of content inclusion and
noninclusion, and as data a theory of content-preserving entailment must
comply with:
4.465 The logical product of a tautology and a proposition says the same
thing as the proposition. Therefore that product is identical with the
proposition.
Thus, in particular, A says the same as A∧(B∨¬B). If one only looks at truth
sets, these coincide. But it needn’t be the case that t(A) = t(A ∧ (B ∨ ¬B)),
so the two cannot express the same thick proposition. In fact, I think that
the interpretation of TLP sense-containment provided by Negro in terms of
inclusion of falsity-grounds (Negro, 2017, 17-20) does a much better job than
my little 2C semantics in making sense of the whole stance of the TLP with
regards to logical consequence and content inclusion.
However, when it’s about characterizing propositional content and same-
saying, it’s the thick proposition view that gets it right, not the merely
truth-conditional view – or so I claim. How can ‘Jane fell out of bed’ say the
same as ‘Jane fell out of bed and either John is 6 feet tall or not’ ? Only the
latter is, partly, about John’s heights. Mary may have brought it about that
13
Jane fell out of bed without bringing it about that Jane fell out of bed and
either John is 6 feet tall or not (Perry, 1989).
It still seems to me reasonable to say, at least, that the direction into
which truth-conditional semantics has de facto been developed into SPWS
and UCLA thin propositions is not straightforwardly mandated by the TLP
– nor are, therefore, its problems. One should not be puzzled by the puzzle-
ment of authoritative interpreters like Ramsey or Black: of course they had a
hard time coordinating what the TLP says about sense containment in 5.122
with the direction it takes elsewhere. I’m more puzzled by the fact that the
authoritative interpreters already saw the merely truth-conditional direction
as the default one, and treated the other one, the content-containment direc-
tion, as the one which shouldn’t have been there, or should have been treated
as a mere metaphor.
References
R.B. Angell. Three systems of first degree entailment. Journal of Symbolic
Logic, 47:147, 1977.
14
K. Fine. Truthmaker semantics for intuitionistic logic. Journal of Philosoph-
ical Logic, 43:549–77, 2014.
W.T. Parry. Ein axiomensystem fr eine neue art von implikation (analitische
implikation). Ergebnisse eines Mathematischen Kolloquiums, 4:5–6, 1933.
15
R.M. White. Wittgenstein’s Tractatus Logico-Philosophicus: A Reader’s
Guide. Continuum, London, 2006.
16